U.S. Pat. Nos. 5,119,436 and 6,869,869 relate to wire bonding systems and associated vision systems, and are hereby incorporated by reference in their entirety.
In the processing and packaging of semiconductor devices, teaching operations using vision systems are often utilized. For example, before a wire bonding operation is performed on a batch of semiconductor devices (e.g., devices such as a semiconductor die mounted on a leadframe), it is typically desired to “teach” an eyepoint (or multiple eyepoints) of a sample device. By “teaching” the sample device, certain physical data related to the sample device is stored (e.g., in the memory of a wire bonding machine). This physical data is used as a reference during processing of the batch of devices, for example, to ensure proper positioning or alignment of each of the batch of semiconductor devices to be processed (e.g., to be wire bonded).
Thus, in the context of a wire bonding operation, a wire bonding machine uses a vision system (e.g., a Pattern Recognition System or PRS) to find a previously taught pattern (e.g., an eyepoint, a fiducial, etc.) for aligning a semiconductor device after it is presented at the bond site and before the wires are bonded (e.g., before the wires are bonded between the semiconductor device and a leadframe supporting the semiconductor device). Traditionally, an eyepoint is taught on the wire bonding machine based on a sample device where an operator targets an area on the sample device with a teach window. Certain conventional techniques (e.g., algorithms) are used in conjunction with a vision system to scan the targeted eyepoint.
One conventional teaching technique relates to scanning the sample device (e.g., a selected portion of the sample device) using a normalized grayscale correlation system (i.e., NGCS). Through such a technique, grayscale values are assigned based on what a vision system detects is present at each location. For example, when a bond pad of a semiconductor device is scanned grayscale values are assigned to the scanned location. After the desired region is scanned, a library of grayscale values (associated with corresponding scanned positions) is stored. When the actual semiconductor devices of this type are to be wire bonded, the vision system detects the grayscale values at each of the scanned locations and compares these grayscale values to those stored in the library during the teaching process.
Another conventional teaching technique relates to scanning the sample device (e.g., a selected portion of the sample device) and detecting individual edges defined within the scanned region (i.e., edge-based pattern matching). Through such a technique, values are defined based on what a vision system detects is present at each location. For example, when a bond pad of a semiconductor device is scanned an edge value is assigned to the scanned location. After the desired region is scanned, a library of edge values (associated with corresponding scanned positions) is stored. When the actual semiconductor devices of this type are to be wire bonded, the vision system detects the edge values at each of the scanned locations and compares these values to those stored in the library during the teaching process.
Using either of these conventional methods, a weighted score is given to each device to be wire bonded, where the score is a function of a comparison of the taught sample device to the actual device to be wire bonded. If the score exceeds a certain threshold value, the device is acceptable and will be processed (e.g., wire bonded); however, if the score is below the threshold value, automatic operation typically does not continue. For example, the operator may be notified of the low score. Further, a subsequent location or eyepoint may be attempted to obtain an acceptable score. Further still, an alternate algorithm or recovery sequence may be attempted.
Unfortunately, there are a number of problems associated with each of these conventional techniques. A practical reality of semiconductor devices is that different devices from the same batch (or different batches) may exhibit different visual properties even though they are considered to be the same device and have the same electrical functional properties. For example, the surface color or texture may vary from device to device. Such variations may arise due to slightly different fabrication processes used by the different suppliers of the same device. Such variations often result in devices that appear quite different from each other in terms of contrast and reflectivity (e.g., non-linear variations in reflectivity). Thus, the weighted scores that result from conventional pattern matching techniques may be lower because of such variations (e.g., due to the differences in reflectivity of the devices to be wirebonded when compared to the sample taught device). Therefore, although scores are typically used to omit false finds, conventional pattern matching techniques (e.g., NGCS systems, edge-based pattern matching systems, etc.) tend to result in scores below the threshold value even if the device is acceptable for further processing. Another problem resulting from such variations amongst semiconductor devices (e.g., surface variations among the devices) may be an undesirably low mean time between assists MTBA), leading to lower productivity of the automatic wire bonding equipment.
Thus, it would be desirable to provide improved methods of teaching eyepoints for semiconductor device processing, and improved methods of processing semiconductor devices using the eyepoints.